Droidrun Is Now Mobilerun: Phones for AI Agents, Explained
Droidrun is an open-source framework that lets AI agents control Android phones and iPhones with natural language, and the German company behind it now sells the product as Mobilerun. The framework, MIT licensed with about 10,000 stars on GitHub, reads a phone’s screen as structured text an AI model can understand, then taps, swipes and types to finish a task. On top of it, Mobilerun Cloud rents agents hosted Android phones and real iPhones, with phone numbers for verification texts, email inboxes and an MCP server for Claude and Cursor. The company is still Droidrun GmbH, which is why older coverage and the GitHub organization use the Droidrun name.
The quick answer: Droidrun, now Mobilerun, is one of the most established names in phone use for AI agents, with a large open-source community, a peer-reviewed research paper, and a cloud service that goes from a free plan to enterprise fleets. Its hosted product leans hard into realistic device identity, so its acceptable use rules are worth reading closely. A disclosure: we make Carly, an AI assistant that makes and answers calls and texts, which is a different job from operating a phone’s apps. Everything below comes from Mobilerun’s site, docs and GitHub, plus funding coverage.
Droidrun at a glance
Checked against Mobilerun’s site, documentation, pricing and GitHub repositories on November 12, 2026.
| Droidrun / Mobilerun | |
|---|---|
| What it is | An open-source framework for AI agents on phones, plus a cloud of phones to run them on |
| Open source | The framework is MIT licensed, about 10,000 GitHub stars and 1,000+ forks |
| Devices | Your own Android or iPhone, emulators, hosted Android phones and real iPhones |
| AI models | OpenAI, Anthropic, Gemini, xAI, DeepSeek, Ollama, OpenRouter and compatible models |
| Cloud plans | Free, then $20, $99 and $299 a month, plus Enterprise |
| Company | Droidrun GmbH, Osnabrück, Germany, founded April 2025 |
| Best for | Developers building mobile agents, app testing and automation on phones |
Why the name changed
The product’s own documentation puts it simply: “mobilerun is the product; Droidrun is the company, which is why coverage from 2025 says Droidrun.” The GitHub repository that started as Droidrun now lives at droidrun/mobilerun, and the old docs address redirects to Mobilerun’s. The new name also fits a product that has grown past Android: the framework now drives iPhones over USB and iOS simulators too.
The switch happened in 2026. The mobilerun package first appeared on PyPI in April 2026, and the old droidrun package is now a compatibility shim whose description reads “droidrun has been renamed to mobilerun.” New installs use pip install mobilerun or uv tool install mobilerun.
What the framework does
The open-source framework is the core of everything. Its README describes it as “an open-source framework for mobile app testing, QA, and running AI agents on real Android and iOS devices.” You give it a task in plain language, such as “Open settings and turn on dark mode,” and it works through the phone’s apps to get there.
How it works:
- Reading the screen. A small companion app called the Portal turns on Android’s accessibility service and exposes the screen’s interface as structured data. When Droidrun raised its pre-seed round, Tech.eu described it as infrastructure “that turns app interfaces into structured text for LLMs.” The framework can combine that tree with screenshots when it needs visual context.
- Any model. It works with OpenAI, Anthropic, Gemini, xAI, Ollama, DeepSeek, OpenRouter and any OpenAI-compatible model, so you can run local models or frontier ones.
- Direct or reasoning mode. Simple tasks run directly; a reasoning mode plans complex, multi-step workflows.
- Many entry points. You can run it from the command line, a terminal UI, Docker or Python code, extend it with custom tools and structured output, and trace runs with Arize Phoenix or Langfuse.
- App cards. Short notes about how specific apps work, which help the agent navigate them.
The app cards idea comes from research. Co-founder Niels Schmidt is a co-author on a paper on curiosity-driven knowledge retrieval for mobile agents, evaluated on the AndroidWorld benchmark with the open-source framework and published at the ACM Web Conference 2026.
Supported devices
The framework runs locally on your own hardware:
| Target | How it connects |
|---|---|
| Android phone | ADB or the Portal app, Android 10 and later |
| Android emulator | A running emulator over ADB |
| iPhone | The mobilerun-ios bridge over USB |
| iOS simulator | The iOS Portal app |
| Mobilerun Cloud | Hosted Android phones and iPhones, or your own phone connected to the cloud |
You need Python 3.11 to 3.13 and, for Android, a device with USB debugging turned on.
Mobilerun Cloud
For teams that don’t want to manage phones, Mobilerun Cloud is the commercial product built on the framework. Its site describes it as “cloud phones for AI agents.” What it adds:
- Hosted phones. On-demand Android phones on every plan, dedicated Android phones with their own residential IP address, and real, physical iPhones reserved for you. Stopped phones keep their apps and data.
- Your own phone. For $5 a month, you can control your own Android or iPhone from the cloud.
- Phone numbers. Receive-only SMS numbers for verification codes, priced by country, or bring your own eSIM. Sending texts is Enterprise only.
- Agent Mail. Permanent inbound email addresses your agent can read through the API, including pulling out one-time codes.
- Hosted agents. One API call per task, with a choice of models, billed from credits. A no-code Virtual Assistant, in beta, lets you type a task and schedule it.
- Integrations. A hosted MCP server for Claude Code, Claude Desktop, Cursor and VS Code, plus n8n, Zapier and OpenClaw integrations.
- Observability. A live view of the phone, screenshots, the accessibility tree, step traces and webhooks for every run.
Device identity and networks
Mobilerun’s cloud is built to make each phone look consistent and real to the apps it uses, and this is where it differs most from a plain testing service.
Each phone keeps persistent settings, including model, serial number, IMEI, Android ID, advertising ID and carrier. GPS location, locale, language and time zone can be set for any country. Network traffic runs through Mobilerun Connect, the company’s own network, with residential traffic on every plan and dedicated residential IPs in 21 countries. The docs say the agent “taps, types and swipes with natural timing for realistic app tests.”
Mobilerun’s own summary of acceptable use says the service is for “your own accounts, your clients’ accounts with their permission, your own apps, and public data,” and “not for mass account creation, impersonation, or evading a platform’s terms.” Its terms of service are more specific. They prohibit “any use of proxies to circumvent technical protection measures (including geo-blocking/rate limiting/captcha),” and make the customer solely responsible for not violating an app’s terms, giving “a prohibition of automated use of Third-Party Apps” as an example. Many apps restrict automation, so the identity and network features are best used within those limits.
Plans
Mobilerun’s pricing has four self-serve plans:
| Plan | Price | Includes each month |
|---|---|---|
| Free | $0 | 60 device minutes, $5 of credits, 1 concurrent phone |
| Developer | $20 a month | 600 device minutes, $15 of credits, 5 concurrent phones |
| Startup | $99 a month | 3,000 device minutes, $40 of credits, 25 concurrent phones |
| Scaleup | $299 a month | 9,000 device minutes, $100 of credits, 100 concurrent phones |
| Enterprise | Custom | Unlimited phones, outbound SMS, private fleets and custom SLAs |
Paid plans continue past their allowance at $0.03 per device minute, up to a monthly overage limit, while the free plan simply stops. Credits pay for AI model usage by agents and the assistant. Dedicated phones are add-ons: an Android phone with its own residential IP is $80 a month, or $64 billed annually, and a real iPhone with a dedicated IP is $200 a month, or $160 billed annually.
Benchmarks
Mobilerun publishes benchmark results with every run public. It reports a 98.34% mean official score across 133 iOSWorld tasks and an official score of 1.0 on all 116 AndroidWorld tasks it ran, with replays of each task on its benchmark page. The company notes that its iOSWorld run “is not a leaderboard submission” and used its own setup, including its Mobile Harness, app cards and shared memory, so the numbers aren’t directly comparable with the public leaderboard. They’re the company’s own results.
The company
Droidrun GmbH is based in Osnabrück, Germany, and was founded in April 2025 by Christian Ninstel, its CEO, and Niels Schmidt. In July 2025 it raised a €2.1 million pre-seed round led by Merantix Capital, with SixtyDegree Capital, TKM and May Ventures. The framework reached the #1 spot of the day on Product Hunt under the Droidrun name, and the company says its Discord community has more than 1,800 members.
As a German company, Droidrun falls under EU data protection law. Its docs say Mobilerun isn’t yet SOC 2 or HIPAA certified, with both in progress, and that account data is deleted within 90 days of an account closing unless the law requires it to be kept.
How Droidrun compares
Droidrun is the open-source anchor of the phone-use category. PhoneUse offers cloud Android phones by the minute and robot-tapped iPhones through an MCP server, Glasshand rents dedicated physical iPhones and Macs by the month, and Phonebox sells persistent cloud Android phones billed by the second. Droidrun’s difference is that you can start for free on your own phone with the open-source framework and move to its cloud later, keeping the same agent. Its cloud also bundles more around the phone, including numbers, email inboxes and residential networking.
If the job is phone calls and texts rather than an agent working through apps, Carly does phone and SMS out of the box: it makes and answers calls and texts from a real business number and also handles email and calendars, from inside ChatGPT or Claude or by text.
Who Droidrun suits
Droidrun is a good fit if:
- You want an open-source mobile agent framework you can run on your own phone today.
- You’re building or testing mobile apps and want an AI agent to work through them.
- You want to choose your own AI model, including local ones.
- You need a path from a laptop prototype to a fleet of hosted phones.
Look elsewhere if:
- You only need a simple phone integration, like sending texts, which other tools handle more directly.
- Your plan relies on creating many accounts or getting around an app’s terms, which its rules forbid.
- You need SOC 2 or HIPAA certification today.
FAQ
What is Droidrun?
Droidrun is an open-source framework that lets AI agents control Android phones and iPhones with natural language. It’s made by Droidrun GmbH in Germany and is now sold as Mobilerun.
Is Droidrun the same as Mobilerun?
Yes. Mobilerun is the product name and Droidrun is the company. The GitHub repository moved to droidrun/mobilerun, and the package installs as mobilerun.
Is Droidrun free?
The framework is free and MIT licensed. Mobilerun Cloud has a free plan with 60 device minutes a month, and paid plans start at $20 a month.
Does Droidrun work with iPhones?
Yes. The framework drives iPhones over USB and iOS simulators, and Mobilerun Cloud offers real, physical iPhones as an add-on.
Which AI models does Droidrun support?
OpenAI, Anthropic, Gemini, xAI, DeepSeek, Ollama, OpenRouter and other OpenAI-compatible models.
Who funds Droidrun?
Droidrun raised a €2.1 million pre-seed round in July 2025, led by Merantix Capital with SixtyDegree Capital, TKM and May Ventures.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
